def test_token_classification_viz_on_true_classes_value_error():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer(explainer_string)
true_classes = ["None", "Location", "None"]
with pytest.raises(ValueError):
ner_explainer.visualize(true_classes=true_classes)
def test_token_classification_explainer_str():
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
s = "TokenClassificationExplainer("
s += f"\n\tmodel={DISTILBERT_MODEL.__class__.__name__},"
s += f"\n\ttokenizer={DISTILBERT_TOKENIZER.__class__.__name__},"
s += "\n\tattribution_type='lig',"
s += ")"
assert s == ner_explainer.__str__()
def test_token_classification_encode():
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
_input = "this is a sample of text to be encoded"
tokens = ner_explainer.encode(_input)
assert isinstance(tokens, list)
assert tokens[0] != ner_explainer.cls_token_id
assert tokens[-1] != ner_explainer.sep_token_id
assert len(tokens) >= len(_input.split(" "))
def test_token_classification_decode():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
input_ids, _, _ = ner_explainer._make_input_reference_pair(
explainer_string)
decoded = ner_explainer.decode(input_ids)
assert decoded[0] == ner_explainer.tokenizer.cls_token
assert decoded[-1] == ner_explainer.tokenizer.sep_token
assert " ".join(decoded[1:-1]) == explainer_string
def test_token_classification_predicted_class_names():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer._run(explainer_string)
ground_truths = ["O", "O", "O", "B-LOC", "O", "O", "O", "O"]
assert len(ground_truths) == len(ner_explainer.predicted_class_names)
for i, class_id in enumerate(ner_explainer.predicted_class_names):
assert ground_truths[i] == class_id
def test_token_classification_predicted_class_names_no_id2label_defaults_idx():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer.id2label = {"test": "value"}
ner_explainer._run(explainer_string)
class_labels = list(range(9))
assert len(ner_explainer.predicted_class_names) == 8
for class_name in ner_explainer.predicted_class_names:
assert class_name in class_labels
def test_token_classification_explainer_init_distilbert():
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
assert ner_explainer.attribution_type == "lig"
assert ner_explainer.label2id == DISTILBERT_MODEL.config.label2id
assert ner_explainer.id2label == DISTILBERT_MODEL.config.id2label
assert ner_explainer.attributions is None
def test_token_classification_explainer_init_attribution_type_error():
with pytest.raises(AttributionTypeNotSupportedError):
TokenClassificationExplainer(
DISTILBERT_MODEL,
DISTILBERT_TOKENIZER,
attribution_type="UNSUPPORTED",
)
def test_token_classification_explain_position_embeddings():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(BERT_MODEL, BERT_TOKENIZER)
pos_attributions = ner_explainer(explainer_string, embedding_type=1)
word_attributions = ner_explainer(explainer_string, embedding_type=0)
for token in ner_explainer.word_attributions.keys():
assert pos_attributions != word_attributions
def test_token_classification_explain_position_embeddings_incorrect_value():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer(explainer_string, embedding_type=0)
incorrect_word_attributions = ner_explainer(explainer_string,
embedding_type=-42)
assert incorrect_word_attributions == word_attributions
def test_token_classification_run_text_given():
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer._run(
"We visited Paris during the weekend")
assert isinstance(word_attributions, dict)
actual_tokens = list(word_attributions.keys())
expected_tokens = [
"[CLS]",
"We",
"visited",
"Paris",
"during",
"the",
"weekend",
"[SEP]",
]
assert actual_tokens == expected_tokens
def test_token_classification_selected_indexes_all():
explainer_string = (
"We visited Paris during the weekend, where Emmanuel Macron lives.")
expected_all_indexes = list(range(15))
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer(explainer_string)
assert len(
ner_explainer._selected_indexes) == ner_explainer.input_ids.shape[1]
for i, index in enumerate(ner_explainer._selected_indexes):
assert i == index
def test_token_classification_selected_indexes_only_ignored_labels():
ignored_labels = ["O", "I-LOC", "B-LOC"]
indexes = [8, 9, 10]
explainer_string = "We visited Paris last weekend, where Emmanuel Macron lives."
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer(explainer_string,
ignored_labels=ignored_labels)
assert len(indexes) == len(ner_explainer._selected_indexes)
for index in ner_explainer._selected_indexes:
assert index in indexes
def test_token_classification_selected_indexes_only_ignored_indexes():
explainer_string = (
"We visited Paris during the weekend, where Emmanuel Macron lives.")
expected_all_indexes = list(range(15))
indexes = [0, 1, 2, 3, 4, 5, 7, 8, 9, 11, 12, 13]
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer(explainer_string,
ignored_indexes=indexes)
assert len(ner_explainer._selected_indexes) == (len(expected_all_indexes) -
len(indexes))
for index in ner_explainer._selected_indexes:
assert index in expected_all_indexes
assert index not in indexes
def test_token_classification_word_attributions():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer(explainer_string)
assert isinstance(ner_explainer.word_attributions, dict)
for token, elements in ner_explainer.word_attributions.items():
assert isinstance(elements, dict)
assert list(elements.keys()) == ["label", "attribution_scores"]
assert isinstance(elements["label"], str)
assert isinstance(elements["attribution_scores"], list)
for score in elements["attribution_scores"]:
assert isinstance(score, tuple)
assert isinstance(score[0], str)
assert isinstance(score[1], float)
def test_token_classification_selected_indexes_ignored_indexes_and_labels():
ignored_labels = ["O", "I-PER", "B-PER"]
ignored_indexes = [4, 5, 6]
explainer_string = "We visited Paris last weekend"
selected_indexes = [
3
] # this models classifies erroniously '[SEP]' as a location
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
word_attributions = ner_explainer(explainer_string,
ignored_indexes=ignored_indexes,
ignored_labels=ignored_labels)
assert len(selected_indexes) == len(ner_explainer._selected_indexes)
for i, index in enumerate(ner_explainer._selected_indexes):
assert selected_indexes[i] == index
def test_token_classification_word_attributions_not_calculated_raises():
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
with pytest.raises(ValueError):
ner_explainer.word_attributions
def test_token_classification_explain_raises_on_input_ids_not_calculated():
with pytest.raises(InputIdsNotCalculatedError):
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer.predicted_class_indexes
def token_classification_custom_steps():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer(explainer_string, n_steps=1)
def test_token_classification_viz():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer(explainer_string)
ner_explainer.visualize()
def token_classification_internal_batch_size():
explainer_string = "We visited Paris during the weekend"
ner_explainer = TokenClassificationExplainer(DISTILBERT_MODEL,
DISTILBERT_TOKENIZER)
ner_explainer(explainer_string, internal_batch_size=1)
